Consultation response system and method

Abstract

To provide a consultation response system for achieving more appropriate consultation.SOLUTION: The consultation response system 1 realizes an input device 12 that receives the content of the consultation from the user, a user feeling estimation unit 111 that estimates the feeling of the user by inputting the content of the consultation to a predetermined machine-learning model, a response generation unit 11,14 that generates a response to the user based on the analyzed feeling of the user and predetermined machine-learning model AI, and an output device 13 that outputs the generated response. The response generation unit inputs the content of the consultation received by the input unit to the predetermined machine learning model to estimate the feeling of the user from the content of the consultation, and generates the response on the basis of the estimated feeling of the user, the learned response method, and the content of the consultation.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a consultation response system and method. [Background technology]

[0002] In consultation response work in the welfare field and other areas, many inquiries are vague, and the appropriate answer to the same question from a user varies depending on the user's psychological state, etc. If a responder who has not earned the user's trust responds to the consultation, they may not give a truthful answer. A skilled responder can close the emotional gap with the user by appropriately showing empathy according to the user and draw out the user's true feelings. However, in recent years, the number of skilled responders has been decreasing, making it difficult to adequately meet the demand for consultations.

[0003] In addition, technologies are known that evaluate whether the user's response is appropriate (Patent Document 1), evaluate whether a caregiver needs support in child-rearing (Patent Document 2), and gain the user's trust and encourage behavioral change (Patent Document 3). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2024-36073 [Patent Document 2] Japanese Patent Publication No. 2022-35959 [Patent Document 3] Japanese Patent Publication No. 2023-66133 Summary of the Invention [Problem to be solved by the invention]

[0005] None of the prior art documents respond to a user's inquiry in accordance with the user's feelings, and therefore cannot provide appropriate advice.

[0006] Therefore, the present invention provides a consultation response system and method that can realize more appropriate consultation. [Means for solving the problem]

[0007] In order to solve the above problem, a consultation response system according to one aspect of the present invention is a consultation response system that responds to a user's consultation, and includes at least one processor and at least one memory used by the processor. The processor executes a predetermined computer program stored in the memory to realize an input unit that accepts input of consultation content from a user, a user emotion estimation unit that inputs the consultation content to a predetermined machine learning model and estimates a user emotion, a response generation unit that generates a response to the user based on the user emotion analyzed by the user emotion estimation unit and the predetermined machine learning model, and an output unit that outputs the generated response to the user. The predetermined machine learning model has been generated by learning a method of responding to the user based on the user emotion estimated from the user's consultation content and a response history for the consultation content. The response generation unit inputs the consultation content accepted by the input unit into the predetermined machine learning model to estimate the user emotion from the consultation content, and generates a response based on the estimated user emotion, the learned response method, and the consultation content. [Effects of the Invention]

[0008] According to the present invention, it is possible to estimate a user's emotions from the content of the user's consultation, generate a response based on the estimated user emotions, the learned response method, and the content of the consultation, and output the generated response to the user. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram illustrating a functional configuration of an information processing system including a consultation response system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram illustrating a hardware configuration of a consultation response system. [Figure 3] FIG. 10 is an explanatory diagram showing how a user's emotions change through a dialogue with the consultation response system. [Figure 4] 10 is a flowchart illustrating a process for generating a consultation response model. [Figure 5] 10A and 10B are diagrams illustrating examples of a learning database and a consultation history database. [Figure 6] 10 is a flowchart showing a response process. [Figure 7] 7 is a flowchart continuing from FIG. 6. [Figure 8] FIG. 10 is a functional configuration diagram of an information processing system including a consultation response system according to a second embodiment. [Figure 9] 10 is a flowchart illustrating a process for generating a consultation response model. [Figure 10] 10A and 10B are diagrams illustrating examples of a learning database and a consultation history database. [Figure 11] 10 is a flowchart showing a response process. [Figure 12] 12 is a flowchart continuing from FIG. 11. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The consultation response system 1 of this embodiment changes the emotions of the user who is seeking advice, and responds in accordance with the response method learned from past consultation history, the change in the user's emotions, and the content of the consultation. In the consultation response system 1 of this embodiment, the response method includes at least one of the order of responses, the timing of responses, and the user's reaction. [Example]

[0011] An embodiment will be described below with reference to Fig. 1 to Fig. 7. Fig. 1 shows the functional configuration of an information processing system including a consultation response system 1. As shown in Fig. 1, the information processing system 1 includes: a consultation response system 1a;

[0012] This information processing system includes, for example, a consultation response system 1, a consultation response model generation system 2, and a user terminal 3 used by a user U. The consultation response model generation system 2 generates a predetermined machine learning model ("AI" in the figure) to be used in the consultation response system 1 and provides it to the consultation response system 1.

[0013] The consultation response system 1 responds to the consultation content input from the user terminal 3, creates a first response, and returns it to the user terminal 3. Furthermore, the consultation response system 1 can also create a second response following the first response and return the second response to the user terminal 3.

[0014] The consultation response system 1 includes, for example, a first response unit 11, an input unit 12, an output unit 13, a second response unit 14, and a switching unit 15.

[0015] The first response unit 11 is responsible for dialogue with the user U and returns a first response according to the content of the user's utterance (content of consultation). The first response unit 11 includes, for example, a user emotion estimation unit 111, a first response generation unit 112, and a consultation history database 113.

[0016] The user emotion estimation unit 111 has a function of estimating the user emotion from the consultation content input by the user U. The method of evaluating the user emotion is not limited.

[0017] The first response generation unit 112, which serves as a "response generation unit," generates a first response using the user emotion analyzed by the user emotion estimation unit 111 and a predetermined machine learning model AI, and outputs the generated first response from the output unit 13 to the user U. The first response is a response for changing the user emotion to a predetermined user emotion. In other words, the first response is a response for opening up the user's heart, and may include, for example, content showing sympathy for the user or content showing praise for the user.

[0018] The consultation history database 113 is a database that manages the history of responses to user consultations. For example, the consultation history database 113 manages, for each series of consultations, the content of utterances made by the user U and the replies sent in response to the content of the utterances. A detailed example of the consultation history database 113 will be described with reference to FIG. 5.

[0019] The input unit 12 has a function of accepting consultation contents from the user U. The input unit 12 receives, as consultation contents, voice uttered by the user U or text input by the user U using a user interface device such as a microphone, a keyboard, or a touch panel, and passes it to the first response unit 11 or the second response unit 14.

[0020] The output unit 13 has a function of providing the first response or the second response to the user U. The output unit 13 provides the first response or the second response to the user U using a user interface device such as a monitor display or a speaker, for example.

[0021] The second response unit 14 can be called a "specialized response unit." The second response unit 14 returns a second response to the user U in accordance with the content of the consultation from the user U. The second response unit 14, for example, includes a plurality of second response generation units 141 as "specialized response generation units," and has a specialized database 142 used by these second response generation units 141. Each second response generation unit 142 is a machine learning model specialized in a respective field of expertise, such as, for example, "childcare," "education," "hobbies," "asset formation," "health," or "human relationships."

[0022] The switching unit 15 has a function of switching communication with the user U from the first response unit 11 to the second response unit 14. Conversely, the switching unit 15 can also switch from the second response unit 14 to the first response unit 11 to make the user U open up again.

[0023] The consultation response model generation system 2 generates a predetermined machine learning model AI. The consultation response model generation system 2 includes, for example, a learning database 21, a machine learning model generation unit 22, and a learning data acquisition unit 23.

[0024] The training database 21 stores training data used to generate a predetermined machine learning model AI. Past consultations and their answers between the counselor C1 and the user U1 are accumulated in the training database 21.

[0025] The machine learning model generation unit 22 generates a predetermined machine learning model AI using the learning data acquired by the learning data acquisition unit 23 from the learning database 21.

[0026] In this embodiment, the consultation response system 1 can automatically accept and respond to a consultation from the user U. However, the consultation response system 1 can also respond to the consultation of the user U via a counselor. In this case, the counselor conveys the response (first response or second response) output from the consultation response system 1 to the user U verbally or in text.

[0027] 2 shows the hardware configuration of the consultation response system 1. Both the consultation response system 1 and the user terminal 3 are basically configured using a computer.

[0028] The consultation response system 1 includes, for example, at least one processor 101, at least one memory 102, a user interface 103, and a communication interface 104.

[0029] The processor 101 is not limited to a CPU (Central Processing Unit), and may include a GPU (Graphics Processing Unit) or an ASIC (Application Specific Integrated Circuit).

[0030] The memory 102 includes a main storage device and an auxiliary storage device. The memory 102 stores a predetermined machine learning model AI, a predetermined computer program (SWP), and the like.

[0031] The user interface (abbreviated as UI in the figure) 103 is a device for exchanging information with the user U or a counselor or system administrator (not shown). The user interface 103 is appropriately equipped with, for example, a monitor display, a keyboard, a touch panel, a speaker, a microphone, etc. The user interface 103 can also exchange information with the user via AR (Augmented Reality) or VR (Virtual Reality).

[0032] The communication interface 104 is a device that communicates with the user terminal 3 via the communication network CN. The type of the communication network CN does not matter.

[0033] The consultation response system can be connected to a storage medium MM. The storage medium MM is configured, for example, as a memory device, a hard disk device, an optical disk device, a magneto-optical disk device, a magnetic tape device, or the like, and non-temporarily stores computer programs and data. The storage medium MM can transfer and store computer programs and data to memory 102. Computer programs and data can also be transferred and stored from memory 1023 to the storage medium MM. A computer program that realizes the functions of the consultation response system 1 can be stored in the storage medium MM, and by connecting the storage medium MM to another computer and installing the computer program stored in the storage medium MM on the other computer, the other computer can be made to function as the consultation response system 1.

[0034] The user terminal 3 includes, for example, a processor 301, a memory 302, a user interface 303, and a communication interface 304. The processor 301, the memory 302, the user interface 303, and the communication interface 304 are basically the same as the processor 101, the memory 102, the user interface 103, and the communication interface 104 described above, and therefore a description thereof will be omitted. However, instead of the predetermined machine learning model AI and the predetermined computer program SWP, the memory 302 stores a predetermined application program App for using the consultation response system 1. The predetermined application program App may be a web browser or a so-called native application.

[0035] The consultation response system 1 can also be connected to an external system 4 via a communication network CN. The external system 4 is, for example, a storage system that stores a predetermined machine learning model AI and / or a specialized database 142. The external system 4 may also be a program server that distributes a predetermined application program App for using the consultation response system 1.

[0036] 3 shows how the user's emotions change through dialogue with the consultation response system 1. As the user moves from the left side to the right side of FIG. 3, the user's feelings change and the user opens up to the consultation response system 1.

[0037] Although it depends on various factors such as the user's nature, personality, and mood of the day, a user who tries to seek advice using the consultation response system 1 is initially defensive and closed off. This initial state (1) is shown on the left side of Figure 3. As shown in state (2) of Figure 3, the user's emotions change when the consultation response system 1 praises or affirms the user.

[0038] State (3) in FIG. 3 shows a state in which the user has opened up and the user emotion has reached a "predetermined user emotion." State (4) in FIG. 3 shows a state in which the user, in an open-minded state, consults with the consultation response system 1. In an open-minded state, the user can continue with professional consultations on childcare, education, tax strategies, etc. In this embodiment, the user's heart is opened up by returning responses including praise, approval, empathy, etc. to the user. Then, in the open-minded state, the system listens to the user's concerns and provides answers.

[0039] 4 shows the process of generating a consultation response model as a predetermined machine learning model. When a counselor C1 responds to a consultation from a user U1, questions, chief complaints, answers, etc. are accumulated in a training database 21. The consultation response model generation system 2 periodically or irregularly learns the consultation case data accumulated in the training database 21 and generates or updates a predetermined machine learning model AI.

[0040] The learning data acquisition unit 23 of the consultation response model generation system 2 accesses the learning database 21 (S101) and determines whether there is unprocessed case data (S102).

[0041] When the learning data acquisition unit 23 determines that unprocessed case data is in the learning database 21 (S102: YES), it acquires the unprocessed case data from the learning database 21 (S103) and passes it to the machine learning model generation unit 22.

[0042] The machine learning model generation unit 22 analyzes the received case data and determines whether the user has opened up (S104). The machine learning model generation unit 22 determines whether the user has opened up (S106), and if it determines that the user has opened up (S106: YES), it sets a predetermined transformation flag in the learning database 21 (S107). Thereafter, the machine learning model generation unit 22 returns to step S101. If it determines that the user has not opened up (S106: NO), the machine learning model generation unit 22 returns to step S101.

[0043] When the machine learning model generation unit 22 determines that there is no unprocessed case data in the learning database 21 (S102: NO), it inputs the questions, answers, chief complaints and final proposals for the chief complaints, and transformation flags in the consultation case to learn a response method (S110). The response method includes, for example, the order in which questions are asked (the order in which questions are asked), the timing of asking questions, and the way of responding.

[0044] 5 shows examples of the learning database 21 and the consultation history database 113. Here, the learning database 21 will be taken as an example for explanation.

[0045] The learning database 21 manages, for example, a management number, a conversation type, a content, and a flag in association with each other. The conversation type indicates the type of conversation. Examples of conversation types include "chief complaint," "question," "answer," and "final proposal."

[0046] The "chief complaint" is the main thing that the user wants to consult about. The question is a question asked by the consultation response system 1 to the user. The answer is the user's response to the question. The "final proposal" is the content that the consultation response system 1 proposes to the user.

[0047] "Content" refers to the content of the chief complaint, question, or answer. Examples of "content" T1 to T16 shown in Figure 5 are shown below. As shown in the following example, the machine learning model AI answers the user's question using a response method that has been learned to be effective in opening up the user. T1: I want to work, but I just can't. T2: I see. By the way, how old are you? T3: N years old. T4: At age N, have you started to feel anxious about the future? T5: Right. T6: May I ask why you can't work? T7: I just can't work hard. I'm a lazy person. T8: Is that so? I think that just coming here to consult with you is a sign of hard work. Is it difficult for you to have confidence? T9: I guess so. T10: Is there anything you can think of that might be causing you to feel unsure? T11: Maybe it's because I did XX as a child and wasn't able to do YY. T12: I see. I'm sorry if I'm wrong, but I think you might not be confident that you can't do your job well if you can't do ZZ. T13: That might be true. T14: Is there anything you would like to try or something you do on a regular basis? T15: It may sound childish, but I've always loved PP and recently I've been wanting to try making one. T16: It's not childish, and I think you're someone who is trying your best. In that situation, why not try to gain some confidence by pursuing your hobbies or doing something you want to do?

[0048] The "flag" is information indicating whether or not there has been a change in the user's emotion. If there has been no change in emotion, the "flag" is set to "none", and if there has been a change in emotion, the "flag" is set to "yes". The value of the flag changes when a change in the user's emotion is detected, and remains the same until another change in the user's emotion is detected. In the example of FIG. 5, when a change in the user's emotion is detected from the 11th answer T11, the "flag" is set to "yes". The value of the "flag" in subsequent conversations remains "yes". The consultation history database 113 is similar to the training database 21.

[0049] 6 is a flowchart showing the response process. The consultation response system 1 of this embodiment generates a first response to open up the user and a second response to respond professionally to the consultation from the open-minded user. That is, the consultation response system 1 of this embodiment causes a change in the user's emotions by repeatedly exchanging first responses with the user, and responds professionally with the second response to the consultation from the user whose emotions have changed.

[0050] The "flag" is reset when the first response process shown in Fig. 6 is started. The consultation response system 1 acquires the user's chief complaint from the input unit 12 (S201). The consultation response system 1 inputs the chief complaint, question, answer, and flag value to the consultation response model AI, and causes the first response generation unit 112 to generate a first response (S202).

[0051] The consultation response system 1 determines whether a first response has been generated (S203), and if the first response cannot be generated (S203: NO), performs error processing (S204). In the error processing S204, the consultation response system 1 outputs, from the output unit 13, a message such as a response, for example, "Uh huh" or "I'm listening," or a consultation end message, such as "I'm sorry. It looks like I can't be of any more help to you."

[0052] When the first response is generated by the first response generating unit 112 (S203: YES), the consultation response system 1 causes the output unit 13 to output the first response (S205).

[0053] The consultation response system 1 waits for a user's response to the first response output in step S205, and acquires the user's response from the input unit 12 (S206). The consultation response system 1 analyzes the acquired user's response and determines whether the user's emotion is a predetermined user emotion (S207).

[0054] If the user emotion is a predetermined user emotion, the consultation response system 1 determines that the user has opened up (S209: YES), and sets the transformation flag in the consultation history database 113 to "yes" (S210).

[0055] After setting the transformation flag, the consultation response system 1 determines whether to proceed to provide a second response that provides a professional answer to the user's concern (S213). If the consultation response system 1 does not proceed to provide a second response (S213: NO), the process returns to step S202.

[0056] When providing a second response (S213: YES), the consultation response system 1 proceeds to step S214 shown in Fig. 7. When the user emotion estimated by the user emotion estimation unit is a predetermined user emotion, the consultation response system 1 generates a response having content different from the content of the response before the predetermined user emotion was detected, based on the estimated user emotion, the learned response method, and the consultation content. When the user emotion estimated by the user emotion estimation unit 111 is a predetermined user emotion, the consultation response system 1 generates a response (second response) having content different from the content of the response (first response) before the predetermined user emotion was detected, based on the estimated user emotion, the learned response method, and the consultation content.

[0057] The consultation response system 1 switches the input unit 12 and the output unit 13 to one of the second response generation units 141 by the switching unit 15 (S214).

[0058] As described above, the second response generation unit 141 is prepared according to the field of expertise, and an appropriate second response generation unit 141 is selected according to, for example, the user's chief complaint. Alternatively, when switching from the first response unit 11 to the second response unit 14, the consultation response system 1 can also allow the user to select the second response generation unit 141.

[0059] The consultation response system 1 outputs a message to the user, such as, "Please select the area you would like to consult about from the following: 1: Child-rearing, 2: Friendships, 3: Inheritance, etc.", and uses the second response generation unit 141 according to the user's selection.

[0060] The second response generation unit 141 inputs the chief complaint, question, answer, and transformation flag thus far into a consultation response model (a predetermined machine learning model AI) to generate a second response such as a question or a final proposal (S215).

[0061] The consultation response system 1 determines whether a second response has been generated (S216). If a second response has not been generated (S216: NO), the consultation response system 1 ends this process. The consultation response system 1 can end this process after notifying the user that the consultation service is ending.

[0062] When the second response is generated (S216: YES), the consultation response system 1 causes the output unit 13 to output the generated second response to the user (S217). Then, the consultation response system 1 acquires a response from the user who has received the second response (S218).

[0063] The consultation response system 1 determines whether to return to providing a first response (S223). The consultation response system 1 inputs the user's response acquired in step S218 into the machine learning model AI to estimate the user's feelings, and if it determines that the user is not open-minded (S223: YES), it can also proceed to the first response process shown in Fig. 6 (S224).

[0064] For example, when the estimated user emotion is determined to be a predetermined negative emotion, the consultation response system 1 can return to the first response process to encourage the user to open up again (S223: YES, S224). Alternatively, the user can instruct the consultation response system 1 to transition to the first response process.

[0065] According to this embodiment configured as described above, it is possible to open up the user's mind and improve the user's sense of familiarity and trust in the consultation response system 1 by using a predetermined machine learning model AI that has learned in the learning database 21 a response method for transforming the user's feelings into a predetermined user feeling. Therefore, the consultation response system 1 can provide the user with a response that will relieve the user's worries and anxieties, improving usability for the user.

[0066] When the user's mood changes to a predetermined one, the consultation response system 1 of this embodiment can transition to a second response process in which a professional response is provided, either automatically or at the user's option. This allows the system to provide a more appropriate response (or suggestion) to the user's concerns, further improving usability for the user. [Example]

[0067] A second embodiment will be described with reference to Figures 8 to 12. In this embodiment, the differences from the first embodiment will be mainly described. The consultation response system 1A of this embodiment determines the change in the user's feelings in more detail than the first embodiment.

[0068] 8 shows the functional configuration of an information processing system including a consultation response system 1 A. A user emotion estimation unit 111A of the consultation response system 1 A includes detection units 1111 and 1112 that detect changes in a plurality of different user emotions.

[0069] The first emotion change detection unit 1111 has a function of detecting a first emotion change. The second emotion change detection unit 1112 has a function of detecting a second emotion change. When a second emotion change is detected after the first emotion change is detected, the consultation response system 1A determines that the user emotion has reached a predetermined user emotion.

[0070] In this embodiment, it is assumed that the first emotion change occurs less frequently than the second emotion change, and that the second emotion change is closer to the user's open-minded state than the first emotion change. For example, the first emotion change is "affinity," and the second emotion change is "trust (or trust relationship, trust)." After the user feels a sense of affinity with the consultation response system 1A, the user comes to trust the consultation response system 1A and opens up. In this way, the consultation response system 1A of this embodiment assumes that the user gradually opens up, and switches to providing the second response when the user opens up.

[0071] Figure 9 shows the process of generating a consultation response model AI. The following explains the differences from the process described in Figure 4.

[0072] The learning data acquisition unit 23 of the consultation response model generation system 2A acquires unprocessed case data from the learning database 21A (S103) and passes it to the machine learning model generation unit 22A.

[0073] When the machine learning model generation unit 22A acquires the received case data (S103), it detects whether the case data includes a first emotional transformation (S104A), and then detects whether the case data includes a second emotional transformation (S105A).

[0074] When information whose appearance frequency is less than a predetermined first threshold Th1 is detected in the content of the user's consultation, the user feeling deduction unit 111A deduces that the user has reached a first emotion change (S106A: YES). When the user feeling deduction unit 111A detects the first emotion change, it sets a first flag in the learning database 21A described in FIG. 10 to “yes.” When the user feeling deduction unit 111A cannot detect the first emotion change (S106A: NO), it skips step S107A and proceeds to step S108A.

[0075] The user feeling deduction unit 111A deduces that a second emotion change has been detected when information less than a predetermined second threshold Th2, which is smaller than a predetermined first threshold Th1 for detecting a first emotion change, is detected in the content of the user's consultation (S108A: YES).When a second emotion change is detected, the user feeling deduction unit 111A sets a second flag in the training database 21A to “YES” (S109).

[0076] When the machine learning model generation unit 22A determines that there is no unprocessed case data in the learning database 21A (S102: NO), it inputs the questions, answers, chief complaints and final proposals for the chief complaints, and transformation flags in the consultation case to learn a response method (S110). As described above, the response method includes, for example, the order in which questions are asked (the order in which questions are asked), the timing of asking questions, and the way of responding.

[0077] 10 shows examples of the learning database 21A and the consultation history database 113A in this embodiment. Here, the learning database 21A will be taken as an example for explanation.

[0078] As described in Fig. 5, the learning database 21A manages, for example, a management number, a conversation type, a content, and a flag in association with each other. The database 21A in Fig. 9 differs from the database 21 in Fig. 5 in that the flags include a first flag and a second flag.

[0079] The first flag is information indicating whether or not a first emotion change has been detected. If a first emotion change has been detected, the first flag is set to "Yes." If a first emotion change has not been detected, the first flag is set to "No."

[0080] The second flag is information indicating whether or not a second emotion change has been detected. If a second emotion change has been detected, the second flag is set to "Yes." If a second emotion change has not been detected, the second flag is set to "No."

[0081] In the example of Figure 10, the first emotional change is detected in the user's response T11, "Maybe it's because I did XX as a child and couldn't do YY, and it's still affecting me," and the first flag is set to "Yes." As a result of continuing the dialogue, the second emotional change is detected in the user's response T15, "It may be childish, but I've always liked PP, and recently I've been wanting to try making one," and the second flag is set to "Yes."

[0082] As described above, information whose frequency of occurrence is less than a predetermined first threshold Th1 was detected from the user's answer T11, so a first emotion change was detected in the answer T11. Furthermore, information whose frequency of occurrence is less than a predetermined second threshold Th2, which is smaller than the first threshold Th1, was detected from the user's answer T15, so it is estimated that a second emotion change was detected in the answer T15.

[0083] As the user opens up and becomes more comfortable with the consultation answering system 1A, an answer closer to the user's true feelings can be obtained. Answers closer to the user's true feelings are less likely to be spoken openly, and therefore are thought to appear less frequently in the consultation history. Therefore, the consultation answering system 1A of this embodiment estimates the user's open-mindedness in multiple stages based on the frequency of appearance of information included in the user's answers.

[0084] Fig. 11 is a flowchart showing the response process. As described in the first embodiment, the consultation response system 1A generates a first response for opening up the user and a second response for professionally responding to the consultation from the open-minded user. The process of Fig. 11 will be described, focusing on the differences from the process shown in Fig. 6.

[0085] After resetting the first flag and the second flag, the consultation response system 1A acquires the user's chief complaint (S201), and inputs the chief complaint, question, answer, and flag values ​​to the consultation response model AI to generate a first response (S202).

[0086] The consultation response system 1A determines whether a first response has been generated (S203), and if the first response has not been generated (S203: NO), it performs error processing (S204).

[0087] If a first response is generated (S203: YES), the consultation response system 1A causes the output unit 13 to output the first response (S205).

[0088] The consultation answering system 1A acquires the user's answer to the first answer output in step S205 (S206). The consultation answering system 1A analyzes the user's answer acquired in step S206 and determines whether or not a first emotional change has occurred (S207A). Furthermore, the consultation answering system 1A analyzes the user's answer acquired in step S206 and determines whether or not a second emotional change has occurred (S208A).

[0089] As described above, when the information less than the first threshold Th1 is included in the user's answer, the consultation response system 1A determines that the first emotional transformation is included in the answer. When the information less than the second threshold Th2 (Th2 < Th1), which is smaller than the first threshold Th1, is included in the user's answer, the consultation response system 1A determines that the second emotional transformation is included in the answer.

[0090] When the first emotional transformation is detected in the answer from the user (S209A: YES), the consultation response system 1A sets "yes" in the first flag of the consultation history database 113A (S210A). When the second emotional transformation is detected in the answer from the user (S211A: YES), the consultation response system 1A sets "yes" in the second flag of the consultation history database 113A (S212A).

[0091] If the first emotional transformation is not detected (S209A: NO), or if the second emotional transformation is not detected (S211A: NO), the process returns to step S202.

[0092] The consultation response system 1A determines whether to proceed to the provision of the second response (S213). If it is determined to proceed (S213: YES), it proceeds to step S214 shown in FIG. 12.

[0093] The consultation response system 1A switches the input unit 12 and the output unit 13 to any one of the second response generation units 141 by the switching unit 15 (S二百一十四). The selected second response generation unit 141 inputs the main complaint, question, answer, and transformation flag so far into the consultation response model (a predetermined machine learning model AI) to generate a second response (S215).

[0094] The consultation response system 1A determines whether the second response has been generated (S216). If the second response has not been generated (S216: NO), the process ends.

[0095] When the second response is generated (S216: YES), the consultation response system 1A causes the output unit 13 to output the generated second response to the user (S217). Then, the consultation response system 1A acquires a response from the user who has received the second response (S218).

[0096] The consultation response system 1A determines whether a first emotion change has occurred (S219), and further determines whether a second emotion change has occurred (S220). If the first emotion change has been detected (S221: YES), the consultation response system 1A further determines whether a second emotion change has also been detected (S222: YES). The consultation response system 1 determines whether to return to providing a first response (S223), and if it determines to return to providing a first response (S223: YES), it returns to the first response processing shown in FIG. 11 (S224).

[0097] This embodiment configured as described above also has the same effects as those of Embodiment 1. Furthermore, the consultation response system 1A of this embodiment detects changes in the user's emotions at multiple stages and determines whether a predetermined user emotion has been reached, so that it can further draw out the user's true feelings and provide an appropriate response.

[0098] In this embodiment, the first emotional change and the second emotional change are detected, so the response when the first emotional change is detected can be different from the response when the second emotional change is detected in addition to the first emotional change, allowing for a more appropriate response to inquiries from users.

[0099] The present invention is not limited to the above-described embodiments and includes various modifications. The above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those including all of the described configurations. Furthermore, part of the configuration of one embodiment can be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment can be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment can be added to, deleted from, or replaced with other configurations.

[0100] The above-described configurations, functions, processing units, processing means, etc. may be partly or entirely implemented in hardware, for example, by designing them as integrated circuits. The above-described configurations, functions, etc. may also be implemented in software, with a processor interpreting and executing a program that implements each function. Information such as the programs, tables, and files that implement each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), an IC card, an SD card, a DVD, or other recording media.

[0101] For example, the above embodiment describes the following configuration.

[0102] (Supplementary Note 1) A consultation response system for responding to a user's consultation, comprising at least one processor and at least one memory used by the processor, wherein the processor executes a predetermined computer program stored in the memory to realize an input unit that accepts input of consultation content from a user, a user emotion estimation unit that inputs the consultation content into a predetermined machine learning model and estimates a user emotion, a response generation unit that generates a response to the user based on the user emotion analyzed by the user emotion estimation unit and the predetermined machine learning model, and an output unit that outputs the generated response to the user, wherein the predetermined machine learning model is generated by learning a method of responding to the user based on the user emotion estimated from the user's consultation content and a response history for the consultation content, and the response generation unit inputs the consultation content accepted by the input unit into the predetermined machine learning model to estimate a user emotion from the consultation content, and generates the response based on the estimated user emotion, the learned response method, and the consultation content.

[0103] (Supplementary Note 2) The consultation response system according to Supplementary Note 1, wherein the response method includes at least one of the order of responses, timing of responses, and user reaction.

[0104] (Appendix 3) The consultation response system described in Appendix 1 or 2, wherein, when the user emotion estimated by the user emotion estimation unit is a predetermined user emotion, the response generation unit generates a response having content different from the content of the response before the predetermined user emotion was detected, based on the estimated user emotion, the learned response method, and the consultation content.

[0105] (Supplementary Note 4) The consultation response system according to any one of Supplementary Notes 1 to 3, wherein the user emotion estimation unit estimates that the predetermined user emotion has been detected when the user emotion reaches a first emotion transformation.

[0106] (Appendix 5) The consultation response system according to any one of Appendices 1 to 4, wherein the user emotion estimation unit estimates that the first emotion transformation has occurred when information whose occurrence frequency is less than a predetermined first threshold is detected in the consultation content.

[0107] (Appendix 6) The consultation response system described in any one of Appendices 1 to 5, wherein the user emotion estimation unit estimates that a second emotion change has been detected when information less than a predetermined second threshold that is smaller than a predetermined first threshold for detecting a first emotion change is detected in the consultation content, and estimates that the predetermined user emotion has been detected when the second emotion change is detected after the first emotion change has been detected.

[0108] (Appendix 7) A consultation response system described in any one of Appendices 1-6, in which a response different from the content of the response before the specified user emotion was detected is generated by a specialized response generation unit according to the consultation content.

[0109] (Appendix 8) A consultation response method for responding to a user's consultation by a consultation response system that responds to the user's consultation, the consultation response method comprising: accepting input of consultation content from a user; inputting the consultation content into a predetermined machine learning model to analyze user emotions; the predetermined machine learning model being generated by learning a method of responding to the user based on user emotions estimated from the user's consultation content and a response history for the consultation content; inputting the accepted consultation content into the predetermined machine learning model to estimate user emotions from the consultation content; generating a response based on the estimated user emotions, the learned response method, and the consultation content; and outputting the generated response to the user.

[0110] (Supplementary Note 9) The consultation response method according to Supplementary Note 8, wherein the response method includes at least one of the order of responses, timing of responses, and user reaction.

[0111] (Appendix 10) A consultation response method as described in Appendix 8 or 9, wherein, when the estimated user emotion is a predetermined user emotion, a response having content different from the content of the response before the predetermined user emotion was detected is generated based on the estimated user emotion, the learned response method, and the consultation content. [Explanation of symbols]

[0112] 1, 1A: consultation response system, 2, 2A: consultation response model generation system, 3: user terminal, 11: first response unit, 12: input unit, 13: output unit, 14: second response unit, 21, 21A, 113, 113A: database, 141: second response generation unit

Claims

1. A consultation response system for responding to a user's consultation, at least one processor and at least one memory for use by said processor; The processor executes a predetermined computer program stored in the memory, an input unit that accepts input of consultation content from a user; a user emotion estimation unit that inputs the consultation content into a predetermined machine learning model and estimates a user emotion; a response generation unit that generates a response to the user based on the user emotion analyzed by the user emotion estimation unit and the predetermined machine learning model; an output unit that outputs the generated response to the user, the predetermined machine learning model is generated by learning a method of responding to a user based on a user emotion estimated from the consultation content of the user and a response history to the consultation content; The response generation unit inputs the consultation content received by the input unit into the predetermined machine learning model, estimates a user's emotion from the consultation content, and generates the response based on the estimated user's emotion, the learned response method, and the consultation content. Consultation response system.

2. The response method includes at least one of the order of responses, timing of responses, and user reaction. The consultation response system according to claim 1.

3. When the user emotion estimated by the user emotion estimation unit is a predetermined user emotion, the response generation unit generates a response having content different from a content of a response before the predetermined user emotion was detected, based on the estimated user emotion, the learned response method, and the consultation content. The consultation response system according to claim 2.

4. The user emotion estimation unit estimates that the predetermined user emotion has been detected when the user emotion reaches a first emotion transformation. The consultation response system according to claim 3.

5. The user emotion estimation unit estimates that the first emotion change has occurred when information having an appearance frequency less than a predetermined first threshold is detected in the consultation content. The consultation response system according to claim 4.

6. The user emotion estimation unit estimates that a second emotion change has been detected when information less than a predetermined second threshold, which is smaller than a predetermined first threshold for detecting a first emotion change, is detected in the consultation content, and estimates that the predetermined user emotion has been detected when the second emotion change is detected after the first emotion change has been detected. The consultation response system according to claim 3.

7. A response with a content different from the content of the response before the predetermined user emotion is detected is generated by a specialized response generation unit according to the consultation content. The consultation response system according to claim 3.

8. A consultation response method for responding to a user's consultation by a consultation response system for responding to a user's consultation, comprising: Accepts input of consultation details from users, inputting the consultation content into a predetermined machine learning model to analyze user sentiment; the predetermined machine learning model is generated by learning a method of responding to a user based on a user emotion estimated from the consultation content of the user and a response history to the consultation content; inputting the received consultation content into the predetermined machine learning model to estimate a user's emotion from the consultation content, and generating a response based on the estimated user's emotion, the learned response method, and the consultation content; Outputting the generated response to the user. Consultation response method.

9. The response method includes at least one of the order of responses, timing of responses, and user reaction. The consultation response method according to claim 8.

10. When the estimated user emotion is a predetermined user emotion, a response having a content different from a content of a response before the predetermined user emotion is detected is generated based on the estimated user emotion, the learned response method, and the consultation content. The consultation response method according to claim 9.

Images